powell-jackson, t; mazumdar, s; mills, a (2015) financial ... · sumit mazumdar institute for human...
TRANSCRIPT
Powell-Jackson, T; Mazumdar, S; Mills, A (2015) Financial incen-tives in health: New evidence from India’s Janani Suraksha Yojana.Journal of health economics, 43. pp. 154-69. ISSN 0167-6296 DOI:https://doi.org/10.1016/j.jhealeco.2015.07.001
Downloaded from: http://researchonline.lshtm.ac.uk/2287480/
DOI: 10.1016/j.jhealeco.2015.07.001
Usage Guidelines
Please refer to usage guidelines at http://researchonline.lshtm.ac.uk/policies.html or alterna-tively contact [email protected].
Available under license: Creative Commons Attribution Non-commercialhttp://creativecommons.org/licenses/by-nc/3.0/
1
Financial Incentives in Health: New Evidence
from India's Janani Suraksha Yojana 1
Timothy Powell-Jackson *
London School of Hygiene and Tropical Medicine
Sumit Mazumdar
Institute for Human Development, New Delhi
Anne Mills
London School of Hygiene and Tropical Medicine
First Version: 20th September 2011
This Version: 15th June 2015
This paper studies the health effects of one of the world’s largest demand-side financial
incentive programmes – India’s Janani Suraksha Yojana. Our difference-in-difference
estimates exploit heterogeneity in the implementation of the financial incentive programme
across districts. We find that cash incentives to women were associated with increased
uptake of maternity services but there is no strong evidence that the JSY was associated
with a reduction in neonatal or early neonatal mortality. The positive effects on utilisation
are larger for less educated and poorer women, and in places where the cash payment was
most generous. We also find evidence of unintended consequences. The financial incentive
programme was associated with a substitution away from private health providers, an
increase in breastfeeding and more pregnancies. These findings demonstrate the potential
for financial incentives to have unanticipated effects that may, in the case of fertility,
undermine the programme’s own objective of reducing mortality.
*Corresponding author: Timothy Powell-Jackson, London School of Hygiene and Tropical Medicine, 15-17
Tavistock Place, London, WC1H 9SH. Email: [email protected]. Tel: +44(0)20 7927
2974.
We thank Arnab Acharya and Marcos Vera-Hernandez for helpful comments on an earlier draft of this article.
We are grateful to Billy Stewart for his guidance and support during this project. This study was funded by
UKaid from the Department for International Development. The views expressed do not necessarily reflect the
Department’s official policies. All errors are our own.
2
I. Introduction
One of the main challenges for global health is to identify policies and strategies that
improve the health of women and children (United Nations, 2010). The traditional focus of
much of the medical literature has been on intervention research resulting in unprecedented
knowledge on what health technologies work (Bhutta et al., 2008; Campbell and Graham,
2006; Jones et al., 2003). Never before have policymakers in developing countries had such
a wealth of evidence at their disposal. Indeed, countries that achieved universal coverage
of life-saving interventions have seen rapid reductions in mortality. For example, over the
past two decades Thailand, Vietnam and Sri Lanka have developed a comprehensive
primary health care system. All these countries between 1990 and 2006 witnessed average
yearly reductions in under five mortality of over 5 percent (Rohde et al., 2008). Yet across
the developing world more broadly there are large gaps in coverage, particularly amongst
the poorest (Bhutta et al., 2010). A key question then is whether there are policies that can
be introduced within health systems – termed here health system interventions – which can
be shown to improve uptake of priority health services.
In an effort to improve population coverage of health interventions and narrow the
differences between income groups, policymakers in developing countries are becoming
increasingly bold in their reforms. One promising strategy is to provide financial incentives
to individuals who exhibit certain behaviours that improve health.2 This is the key feature
of various programmes that have become popular in recent years. Whether the incentive
takes the form of conditional cash transfers, vouchers or one-off cash payments, the central
idea of providing monetary rewards conditional on measurable actions is the same.
Financial incentives have courted considerable controversial, with views ranging from “as
close as you can come to a magic bullet” to a “form of bribery” (Dugger, 2004; Marteau et
al., 2009). Critics point to the theoretical possibility of unintended consequences as well as
moral concerns over their use, particularly in a health setting.
This paper studies the early effects of one of the largest cash incentive programmes for
health in the world. With an annual expenditure of 8.8 billion rupees or $207 million, and
an estimated 7.1 million individual beneficiaries,3 India’s national Janani Suraksha Yojana
(JSY) provides cash to women who give birth in a health facility. The JSY provides an
ideal testing ground to examine the effects of financial incentives on health. Although
officially launched in 2005, implementation of the JSY across districts was incremental,
providing variation in its coverage. At the same time, much of the health policy
environment in India is common within states, which gives us more confidence that district
variation in the JSY is not acting as a proxy for other policy initiatives. A second advantage
of this setting is the narrow focus of the JSY on women at childbirth. This provides greater
scope for examining unintended consequences of the financial incentives on closely related
but non-incentivised behaviours. A third advantage is the scale at which the JSY was
2 In this paper we are interested in demand-side financial incentives, rather than provider payment mechanisms
such as pay-for-performance. The latter reward physicians for improvements in quality of care and other
measures, and are popular in the US and UK. For brevity, we will use the term financial or cash incentives in
health to refer to schemes that target the users of health care. 3 These figures refer to 2007/08, the financial year closest to our study period.
3
implemented. This differentiates our study from carefully controlled small scale (incentive)
experiments, whose external validity has at times been questioned (Deaton, 2010).
We identify the effect of the JSY on health care seeking behaviour and health status by
exploiting the substantial variation in implementation of the JSY across districts. Using
data on women who gave birth between 2001 and 2008 from two rounds of India’s District
Level Health Survey (DLHS), our empirical approach examines whether the JSY can
account for cross-district patterns in health care utilisation and health status over time. In
estimating the effect of the JSY, this difference-in-difference strategy allows us to control
for time invariant unobservables at the district level that influence study outcomes and are
correlated with the expansion of the JSY. Using changes in the intensity of the JSY to
identify programme impacts, nevertheless, gives rise to endogeneity concerns. Early
adopters of the JSY, for example, may have been districts that were highly motivated to
make improvements in maternal health services. While we provide extensive robustness
checks on our main findings, we are unable to rule out the possibility of confounding and
refrain from making strong claims of causality.
Our results show that the JSY was associated with an increase in the proportion of women
who give birth in a public health facility. Estimates suggest the magnitude of this effect
was reasonably modest. The positive association between the JSY and women giving birth
in a public health facility was driven almost entirely by increases in the use of primary
health centres and community health centres, providers offering more basic services than
those available at the district hospital. In addition, we present evidence on the effect of the
JSY on health outcomes, finding no strong evidence of an effect on either neonatal
mortality (deaths within 28 days of birth) or one-day mortality (deaths within 24 hours of
birth). We note, however, that confidence intervals are not sufficiently tight to reject a
modest effect of the JSY on these mortality outcomes.
We also provide evidence on a number of unintended consequences. First, a lack of
implementation of the JSY much beyond the public sector means that the financial
incentives resulted in women substituting away from giving birth in the private sector.
Second, results show that the JSY had a positive, statistically significant effect on
pregnancies. Third, we find evidence of indirect benefits. Women in JSY districts were
more likely to start early breastfeeding within one hour of childbirth.
This paper contributes to the existing literature by reporting more credible treatment effects
than previous studies on the JSY. Our main results are consistent with much of the evidence
emerging from conditional cash transfer programmes and small scale incentive
experiments.4 We also go beyond the typical study of financial incentives in examining
unintended consequences. Similar to the findings from studies in Brazil (Morris et al.,
2004b) and Honduras (Morris et al., 2004a), we document evidence of unintended effects,
which highlight how important it is for policymakers to consider the full range of effects
in the design of financial incentive schemes. More generally we connect to a second
4 The systematic literature reviews on conditional cash transfers (Lagarde et al., 2007) and demand-side
incentives in health (Murray et al., 2014) provide a detailed summary of much of this evidence.
4
literature evaluating the impact of health system interventions and policies. This is a wide
ranging and challenging area of research (Mills et al., 2008), and one in which much of the
existing econometric evidence focuses on the impact of health financing initiatives.5
Given that the JSY remains a high-profile federal health programme in India, the findings
are of relevance to policy. First, they argue for much better administration of the
programme. If disbursement of the JSY cash were improved, the effect on use of formal
health care would be greater than at present. Second, the findings reinforce the growing
sentiment that demand-side intervention by government can be effective in improving
uptake of health services but alone may be insufficient to improve health outcomes.
Strengthening the quality of primary health care and the referral system in India is thus a
critical complementary strategy, as is staggering supply- and demand-side investments over
time such that individuals are encouraged to use services once quality has improved. Third,
the findings suggest that financial incentives may be an imprecise tool for changing health-
related behaviours. They can have unintended health effects, on fertility for example, which
may undermine the programme’s own objectives. Financial incentives must therefore be
used with caution.
The paper is structured as follows. Section II describes the JSY and addresses the
theoretical predictions of its impact on health-related behaviours. Section III describes the
data. Section IV presents the empirical strategy. Section V presents the main econometric
results and includes a discussion of robustness checks. Section VI examines heterogeneity
in the impact of the JSY, and Section VII offers concluding comments.
II. Background
II.A India’s Janani Suraksha Yojana
Despite the long history of well-intentioned family welfare policies and some recent
progress, maternal and child mortality in India remains high. With 72,000 maternal deaths,
no other country accounts for a larger proportion of global mortality (Kassebaum et al.,
2014). Maternal mortality has fallen by 47 percent from 398 deaths per 100,000 live births
in 1997-98 to 178 deaths per 100,000 live births in 2010-12 (Registrar General of India,
2006, 2013). However, the national picture masks enormous differences across states. For
example, Kerala’s maternal mortality rate is almost five times lower than some of the worst
performing northern Indian states (Registrar General of India, 2013). National surveys
show that institutional deliveries have increased modestly over time but a large proportion
of women continue to give birth at home (International Institute for Population Sciences,
1995; International Institute for Population Sciences and Macro International, 2007). Even
when women do reach a health facility to give birth, health workers are often absent
(Chaudhury et al., 2006; Muralidharan et al., 2011) and the quality of care they receive is
low (Das and Hammer, 2006; Das and Hammer, 2007; Das et al., 2008).
5 See, for example, studies on health insurance (Babiarz et al., 2010; Finkelstein et al., 2011; King et al., 2009;
Manning et al., 1987; Thornton et al., 2010; Wagstaff et al., 2009).
5
It is against this background that the federal government launched the National Rural
Health Mission (NRHM) in 2005. Key elements of the programme include large
investments in health infrastructure, the deployment of three quarters of a million newly
created accredited social health activists as frontline health workers in the community,
strategies to stimulate demand for health services, and decentralisation of the health system
(Ministry of Health and Family Welfare, 2005). One of the more high profile components
of the NRHM is the Janani Suraksha Yojana (translated as “Safe Motherhood Scheme”). It
was launched officially in April 2005, with the objective of improving maternal and
neonatal health through the promotion of institutional deliveries.6 It provides a cash
incentive to women who give birth in a public health facility or, in principle, an accredited
private health provider (Ministry of Health and Family Welfare, 2006).
The JSY programme designates Indian states as low performing or high performing,
varying the cash amount to provide greater incentives in the area of higher priority.
Specifically, women in low-performing states are offered 1,400 Rs ($31) in rural areas and
1,000 Rs ($22) in urban areas, and those in high-performing states are given 700 Rs ($16)
in rural areas and 600 Rs ($13) in urban areas.7 To put these amounts in perspective, annual
Gross National Income per capita was $1000 in 2007 and the average amount paid for
delivery care in the public sector was $25 in 2004 (Bonu et al., 2009). The cash payment is
available to all women in the low-performing states; by contrast, it is offered in high-
performing states only to women living in households below the poverty line, belonging to
scheduled castes and tribes, or those who have had two or fewer live births. The policy
stipulates that the cash is to be disbursed to the mother immediately at the institution itself
and within a week of delivery.
To provide incentives for health workers who encourage women to give birth in a formal
care provider, accredited social health workers are offered a cash payment of between 200
Rs ($4) and 600 Rs ($13) for each delivery attended. The JSY also pays 500 Rs ($11) to
women who give birth at home, conditional on less than two living children and a below
the poverty line card, but since this is a direct continuation of the cash assistance provided
under the National Maternity Benefit Scheme, it does not represent an additional incentive
for eligible women to stay at home for delivery.
6 Ethnographic research in the Indian state of Uttar Pradesh casts doubt on the government strategy to encourage
institutional deliveries as a means to improve the health of women. Jeffrey and Jeffrey (2010) argue that the
context surrounding the government provision of health care presents challenges that neither the NRHM nor
the JSY were intended to address. Decades of mistrust of government health services and controversial family
planning programmes have left a credibility gap not easily filled by offering financial incentives and investing
in new infrastructure. In line with a report by Human Rights Watch (2009), they contend that accountability of
government health providers to the population they serve is key and nothing less than “a dismantling of a long-
standing political economy of health care provision” will help to remedy the situation. 7 The low-performing states consist of Bihar, Chattisgarh, Jharkhand, Orissa, Uttar Pradesh, Uttaranchal,
Rajasthan, Madhya Pradesh, Assam and Jammu and Kashmir.
6
II.B Anticipated effects
Consider a financial incentive programme that rewards families in which the woman gives
birth in a health facility.8 Basic economic theory suggests short-term financial incentives
will increase demand for maternal health care. Financial incentives provided to women
seeking care in the public sector only change the relative prices of different care seeking
options and are thus expected to lead to a substitution away from private health providers
and home births (Gertler and Van der Gaag, 1990).
To the extent that public health providers can meet this increase in demand, financial
incentives will increase utilisation of health services. If instead public health providers are
functioning at full capacity or are unable to increase supply in the short-term, financial
incentives will have little impact on utilisation. Moreover, there may be no overall increase
in utilisation if the financial incentives contribute to crowding-out of the private sector.
Whether an increase in utilisation of public health services improves health outcomes is
not clear-cut, and will depend on differences in the clinical quality of care between the
various health care seeking choices. We would expect the narrowest difference in quality
to be between public and private health providers, particularly in terms of clinical as
opposed to interpersonal dimensions of quality.
While the financial rewards provide explicit incentives to use maternal health services,
implicitly they also serve to incentivise pregnancy. This effect may manifest itself in terms
of a reduction in birth spacing or an increase in total lifetime children for women who
otherwise would not have become pregnant. We also anticipate indirect effects as financial
incentives increase women’s exposure to health information. Greater contact with health
staff exposes women to more information on healthy behaviours concerning the mother
and her neonate. Behaviors shown to have an impact on health outcomes include wrapping
the baby within 30 minutes of childbirth, initiating breastfeeding within one hour, and
dressing the cord with antiseptic (Darmstadt et al., 2005).
II.C Evidence on the JSY
There have been a number of studies on the JSY, some of which have collected primary
household data (Hunter et al., 2014). For the most part these have been descriptive,
documenting progress in the implementation of the programme (Devadasan et al., 2008;
Malini et al., 2008; Verma et al., 2010). By contrast, Lim et al (2010) make claims as to
the causal effect of the JSY. Impact estimates are based on three identification strategies:
individual-level matching, a modified before and after design, and a two-period district
level difference-in-difference approach. The main conclusion from the analysis is that the
JSY increased substantially use of maternal health care and reduced neonatal mortality.
The study by Lim et al (2010) has several important limitations. First, the headline results
are based on the matching and modified before and after design, while estimates from the
difference-in-difference analysis are given less emphasis on the basis that they lack power.
Having imprecise estimates from the district level analysis does not provide justification to
8 For a thorough discussion of the economic rationale of conditional cash transfers, see Fiszbein and Schady
(2009).
7
highlight other methods simply because they have more power. Second, individual
matching based on whether women did and did not receive the JSY cash is unlikely to
provide credible estimates of effect because there is reverse causality (women receive the
cash when they give birth in a health facility) and individual unobservables correlated with
outcomes are likely to be important factors in determining who takes up the programme.
The modified before and after study design is also problematic since it must rely again on
the strong assumption of conditional independence (Imbens and Wooldridge, 2009). Third,
a strategy that controls for observables at the district level is more credible because
selection is now at the policy level where it is likely to be based on observed measures of
need. However, the share of births in a district in which the woman received the JSY cash
is an inappropriate measure of treatment because it is mechanically linked to the fraction
of women giving birth in a facility. By definition, it captures not only the availability of the
programme but also demand side factors driving utilisation.
This paper addresses the limitations of past research on the JSY. Given the high profile
nature of the programme, we set out to provide more credible estimates of impact across a
wide range of behaviours. We also provide new findings on how the JSY affects health
seeking choices between different types of provider, the heterogeneity of impacts, and
whether the JSY has unintended consequences.
III. Measures and Data
III.A Study Outcomes
Data on the study outcomes come from the household component of the District Level
Health Survey (DLHS), a repeated cross-section survey designed to provide estimates on
maternal and child health and service utilisation at the district level in India (International
Institute for Population Sciences, 2010). We use data from two rounds of the household
survey. The DLHS-2, conducted over the period 2002-04, interviewed 507,622 currently
married women in 593 districts. The DLHS-3 was carried out in 2007-08 and interviewed
643,944 currently married women in 611 districts.
The married woman questionnaire is modelled closely on India’s established National
Family and Health Survey. It contains measures of health care utilisation and health status
that the JSY would be expected to improve. Our main utilisation outcome is births in a
health facility, measured using information on the place of delivery of the woman’s most
recent birth. The analysis also considers variants on this outcome, such as the type of health
provider chosen, whether a health worker was in attendance and the type of procedure
performed at delivery.
Our main measures of health status are one-day mortality (death of a baby within 24 hours
after being born alive) and neonatal mortality (death of a baby within 28 days after being
8
born alive). Both are measured using information on the birth history of women.9 The
financial year of the most recent delivery and each live birth is established using
information on the year and month reported by women.10 The DLHS-3 limits the recall
period of birth histories to 1st January 2004, while those in DLHS-2 are not truncated.
However, to ensure recall periods are approximately the same in the two survey rounds, we
drop all observations prior to 1st April 2001. Thus, when we stack the data from the two
survey rounds, we have observations in every financial year from 2001/02 to 2007/08.
An important contribution of this paper is to consider the effect of the JSY on a second set
of outcomes that we refer to as unintended consequences of the programme. These are
outcomes that did not feature in the stated objectives of the programme and are in this sense
unintended. They include the likelihood of giving birth in a private health facility, getting
pregnant in a given year, and breastfeeding immediately after childbirth. We establish
whether a woman was pregnant in a given year using the pregnancy histories contained in
the survey. To measure breastfeeding, women were asked if and when they started
breastfeeding the child of their most recent delivery. We focus on breastfeeding within the
first hour, when information from health providers on the benefits of timely breastfeeding
is most likely to take effect. All outcomes in this study are comparable across the two
survey rounds, both in terms of how they are defined and the interview questions used to
elicit the required information.
Summary statistics on the outcome measures before and after the start of the JSY are shown
in Panel A of Table 1. Neonatal mortality fell over the course of the two periods from 33
to 27 deaths per 1,000 live births. Facility births saw a modest increase over time but still
more than half of women continued to give birth at home. Around 8% of women gave birth
by caesarean section and a further 2% had an assisted delivery with forceps or a ventouse,
neither of which changed much over time. By contrast, use of antenatal care and
breastfeeding improved over time. The proportion of women who reported being pregnant
in any given year was 8%. In addition to information on study outcomes, we exploited data
on a broad range of socio-demographic characteristics as detailed in Panel B of Table 1.
The data contain a district identifier which we use to estimate specifications with district
fixed effects. However, because the administrative boundaries of some districts changed in
the period between the two surveys, we map new districts in the DLHS-3 onto their old
counterparts in the DLHS-2 data. In most cases this was possible, leaving 587 districts that
were consistently defined across the two datasets.11 In estimating the effect of the JSY on
care seeking behaviour and health status, for lack of data we assumed that the district in
which women are residing at the time of interview was the same as the one where she gave
9 Unless truncated, a birth history documents every birth a woman has had during her lifetime. It typically
includes the birth outcome, sex of the child, birth order, month and year of childbirth, age of woman at childbirth
and, if the child died, age at death. 10 We work in financial years (1st April to 31st March) throughout because the government’s annual budgetary
cycle is likely to correspond more closely to the introduction of the JSY than calendar years. 11 In cases where the geographical boundaries of newly created districts cut across two or more old districts,
we were unable to map the new districts onto their old counterparts.
9
birth. Available evidence suggests that residents of a district rarely travel to other districts
to seek healthcare.12
III.B JSY Coverage
Our estimation strategy rests on there being variation in the implementation of the JSY. We
exploit such variation at the district level, the administrative unit directly below the Indian
state which has responsibility for planning and implementation of federal and state policies.
If the financial incentives of the JSY are to bite, households should be exposed to
information about the programme and financial incentives should reach eligible women.13
Data on the latter provide the foundation for our measure of JSY penetration and is based
on responses to the question: “Did you receive any government financial assistance for
delivery care under the Janani Suraksha Yojana or state-specific scheme.” Specifically, we
use the term JSY coverage to refer to the number of women who gave birth in a public
facility and received the cash as a proportion of women who gave birth in a public facility.14
Full coverage thus implies every woman giving birth in a public health facility receives the
financial incentive. Because coverage of the JSY is constructed from the sample of women
who delivered in a public facility, it is primarily a supply-driven measure of the intensity
of implementation. It is affected not by the demand for care but rather the government’s
ability to make the programme available to women at the level of service delivery, an
assertion we test below.
Our measure of JSY implementation is based on beneficiary data from households rather
than administrative data (eg. budget releases or district expenditure) for several important
reasons. First, such administrative data may reflect only the intention of the government,
whereas information on whether a district has JSY beneficiaries implies that the
government has taken all the necessary steps to start the programme on the ground. Second,
there is no reason to believe administrative data would be any more reliable than household
data. In fact, such information is easy to manipulate systematically and incentives are likely
to be there to do so.
Table 2 shows the expansion of the JSY programme over time. In its first year, JSY
coverage was less than 10% in 279 of the 587 districts, while only a handful of districts had
coverage over 50%. Over time coverage of the JSY at the district level increased. In the
third year of the programme, JSY coverage was more than 10% in 489 of the 587 districts.
12 In a recent survey of women in Uttar Pradesh, we find that only 1.8% of women giving birth in a facility
travelled outside of their district for delivery. Sood et al (2014) find little evidence of cross-district healthcare
seeking in a study of health insurance in Karnataka. 13 A study carried out in 2008 in the high-focus states of Bihar, Orissa, Uttar Pradesh, Madhya Pradesh and
Rajasthan found that four-fifths of women were aware of the scheme and almost half of women giving birth in
a health facility received the JSY cash (UNFPA, 2009). 14 Due to imprecise wording, this question picked up responses that refer to the National Maternity Benefit
Scheme (MBS), an initiative that preceded the JSY up until its official introduction in April 2005 (see Section
II for more detail). This explains why 7.4 percent of women giving birth in a health facility report receiving a
cash payment in 2004/05, before the JSY was even official government policy. We code the JSY coverage
variable as zero prior to the official start of the programme. While the JSY is not limited to the public sector,
our measure of coverage considers only public sector recipients of the financial incentive because only a few
nonstate health providers – in contrast to all health providers in the public sector – were accredited and able to
participate in the JSY.
10
Figure 1 illustrates well the considerable variation in the expansion of the programme
between districts and over time. It also provides descriptive evidence of the relationship
between the JSY programme and facility births. Districts where the JSY was progressively
better implemented appear to have the largest increases in the proportion of women giving
birth in government health facilities.
In anticipation of the empirical analysis, we recognise that variation in the coverage of the
JSY across districts is unlikely to be random. Discussions with policymakers and other
stakeholders engaged with the JSY suggest that the introduction of the programme was
prioritised in socioeconomically disadvantaged places. At the national level, the JSY was
explicitly prioritised according to high-focus and low-focus states. More importantly,
however, interviews indicated that the JSY was prioritised within states at the district level.
For example, in the state of West Bengal, health sector reforms including the JSY gave
particular attention to six focal districts, identified on the basis of health indicators, poverty
and socially marginalised population groups.15
Empirically we can examine the relationship between JSY coverage and several
socioeconomic variables highlighted by policymakers. In Table A1 of the Appendix, we
run a district-level regression of JSY coverage on poverty incidence, the tribal population
share and average household wealth showing that the three variables of interest are strong
predictors of JSY coverage. Broadly this remains true when we include state fixed effects.
The data support the qualitative evidence in showing the role of these district characteristics
in influencing the decision on where to introduce the JSY. In the final column we see that
the share of births in a government facility in the year before the JSY does not predict
subsequent implementation of the programme, suggesting that our measure of JSY is not
picking up demand side factors influencing utilisation of health services.
IV. Empirical Strategy
Our identification strategy uses a difference-in-difference approach to estimate the impact
of the JSY on our study outcomes. We compare changes over time in health care utilisation
and health status with changes in the intensity of the JSY programme. More precisely, in
our basic specification we run a regression of each outcome on JSY coverage while
controlling for year and district fixed effects. The fixed effects absorb variation due to
common temporal shocks and time-invariant district factors.
To increase the strength of causal inference, we also control for a wide range of potential
confounding factors. Formally, let 𝑦𝑖𝑑𝑡 denote our outcome, a binary measure of service
utilisation or health status for observation 𝑖 in district 𝑑 in year 𝑡. Let JSY𝑑𝑡 denote our
measure of programme coverage in district 𝑑 in year 𝑡. Our specification takes the form:
𝑦𝑖𝑑𝑡 = 𝛽0 + 𝛽1JSY𝑑𝑡 + 𝜗𝑡𝑍𝑑𝛽2 + 𝑋𝑖𝑑𝑡𝛽3 +𝜔𝑑 + 𝜏𝑡 + 𝜀𝑖𝑑𝑡, (1)
15 Scheduled tribes are historically disadvantaged people in India, given explicit recognition in India’s
Constitution.
11
where 𝜔𝑑 and 𝜏𝑡 are district and year fixed effects respectively; 𝑋𝑖𝑑𝑡 is a vector of
individual demographic characteristics including education of the mother, education of the
husband, maternal age, household wealth, the recall period (months between interview and
birth of child) and dummies for (categories) of urban residence, religion, ethnicity, parity,
multiple births and survey round; and 𝑍𝑑 is a vector of district-level characteristics. To
model the effect of the programme flexibly, JSY𝑑𝑡 enters the regressions as dummy
variables that correspond to the following levels of coverage: 10-25%, 25-50% and >50%.
We cluster our standard errors at the district level.
To address several sources of potential confounding, we include interactions between the
year of birth and the share of the district population below the poverty line, the tribal
population share, and the district mean of the household wealth asset score, represented by
the term ϑtZd. Data used to generate these district-level variables come from the DLHS-
3,16 which means we are controlling for differential trends based on 2008 values rather than
actual trends.
As is clear from equation (1), we run regressions of each outcome using individual level
data to make the most of the rich micro dataset at our disposal. This allows us to include
controls for a range of individual demographic characteristics that might affect health care
utilisation and health status. In using individual level data, we note that the unit of
observation differs according to the outcome. Each observation is a delivery (the most
recent only) in the utilisation equations, and a live birth in the mortality equations. In the
analysis of pregnancies, the unit of observation is woman-year but we must rely on data
from the DLHS-3 only (2004-2008) because the DLHS-2 did not collect information on
pregnancy histories.
V. Main Results
V.A Use of Health Care and Mortality
Table 3 presents estimates of the effect of the JSY on various measures of health care
utilisation. Panel A present results from our basic specification which includes district and
year fixed effects. In Panel B, we additionally control for district characteristics and
individual demographics.
Column (1) shows that the JSY was associated with an increase in the percentage of women
giving birth with a health worker in attendance at delivery. Specifically, the likelihood of
giving birth with a health worker was 5.6 percentage points higher in districts with JSY
coverage >50% than districts with coverage <10%. At lower levels of JSY coverage, there
16 Our measure of poverty is constructed using information relating to the government system of identifying
poor households. Specifically, it is based on responses to the question: “Does this household have a below the
poverty line (BPL) card?” Because we are interested in controlling for sources of endogeneity that arise from
government decision making processes, this poverty measure – rather than one measured perhaps more reliably
in terms of household consumption – is particularly appropriate for our purposes.
12
was no significant association. Columns (2) and (3) show the effect of the JSY on health
facility births with the same pattern of results. The point estimates indicate that the
programme at levels of coverage >50% was associated with a 7.5 percentage point increase
in facility births and an 11 percentage point increase in public facility births. Columns (4)
to (6) present the effect of the JSY on utilisation by each type of public health facility.
These results imply that the impact on public health facility births was driven largely by
increases in births at community health centres and primary health centres. By contrast,
district hospitals accounted for only a small proportion of the treatment effect. These
findings suggest an expansion in uptake of delivery care services at public health providers
below the district hospital.
Column (7) shows that the JSY did not have an effect on utilisation of antenatal care
services. The point estimates for three or more antenatal care visits are small and
statistically insignificant and the result holds irrespective of how we define the antenatal
care outcome (result not shown).17 This finding is reassuring for our empirical strategy
because we anticipate no large effect given that the financial incentive in the JSY was not
explicitly tied to the use of antenatal care. It suggests that the JSY treatment indicator is
not simply acting as a proxy for other government policies aimed at strengthening maternal
health services. Further results showing the effect of the JSY on the rate of caesarean
sections and assisted deliveries are reported in Table A4 of the Appendix. The JSY at
coverage levels >50% was associated with a decrease in the caesarean section rate and an
increase in assisted deliveries. The negative impact on the caesarean section rate is most
likely explained by the shift away from the private sector (reported in Section V.C) where
the vast majority of caesarean sections are conducted.18
When we include extensive controls for potential confounders the point estimates remain
essentially the same (Panel B of Table 3). For example, the likelihood of giving birth in a
government health facility is 10 percentage points higher in districts with JSY coverage
>50% than districts with coverage <10%. When we include in the model a single treatment
variable indicating JSY coverage >10%, the findings are qualitatively similar (Table A2).
This model is more akin to an intention-to-treat analysis in the sense that the estimates
reflect better the impact of the JSY irrespective of how well districts implemented the
programme. When we include in the model JSY coverage as a continuous variable the
findings remain qualitatively largely unchanged (Table A3). For example, a 1 percentage
point increase in JSY coverage is associated with a 0.2 percentage point increase in
government facility births.
We next turn to the mortality results (Table 4). The results in column (1) show that there is
no strong evidence the JSY reduced neonatal mortality. None of the coefficients on the JSY
coverage dummies are significant at the 5 percent level. At coverage levels >50% the JSY
is associated with a reduction in neonatal mortality of 3.1 deaths per 1,000 live births which
17 Alternative measures of antenatal care utilisation include the number of antenatal care visits. Using a Poisson
regression we find no effect on the number of antenatal care visits. 18 By contrast, at the first level of referral in the public sector only 18 percent of community health centres offer
caesarean sections and less than 10 percent have blood storage facilities (International Institute for Population
Sciences, 2010).
13
is significant at the 10 percent level. In the specification with the full set of controls, we are
able to reject with 95 percent confidence a negative effect of the JSY larger than 5.9 deaths
per 1,000. In columns (2) to (4) we separate out neonatal mortality into its constituent parts
since we anticipate that if the JSY were to reduce mortality, the effect would be strongest
within the first 24 hours of childbirth when maternity care is provided. Results in column
(2) show a negative effect of the JSY on one-day mortality. There is a slight suggestion of
an effect of the JSY at coverage levels >50%. In the specification with the full set of
controls, we are able to reject with 95 percent confidence a negative effect of the JSY larger
than 4.5 deaths per 1,000. Columns (3) and (4) confirm that there was no effect of the JSY
on later neonatal mortality, which provides some confidence that the findings in column
(2) are not spurious for we would not anticipate maternity care to have a direct effect on
the mortality of the baby after the mother is discharged to go home. The mortality findings
remain similar when we include additional controls (Panel B of Table 3), or replace the
JSY coverage dummies with either a single binary treatment variable of >10% coverage
(Table A2 of the Appendix) or a continuous treatment variable (Table A3 of the Appendix).
These findings suggest that the JSY did not have a large effect on neonatal and one-day
mortality but the confidence intervals leave open the possibility that the JSY had a modest
effect at high levels of programme coverage. Why was the association between the JSY
and mortality, at best, only modest? One possibility is that the effect on utilisation was not
sufficiently large to translate into better health outcomes. A second explanation points to
the poor quality of care in the public sector (Chaturvedi et al., 2014; Hulton et al., 2007;
Nagpal et al., 2015; Stanton et al., 2014), and the fact that the JSY increased uptake of
maternity services at health facilities below the district hospital, which are not equipped to
manage emergency complications at childbirth.
V.B Magnitudes and Simple Cost-Effectiveness
According to our estimates, the JSY encouraged an additional 710 thousand women in India
to give birth in a health facility in 2007/08.19 In a quick calculation using programme
expenditure data, we estimate that the government spent $292 of JSY money for each
additional facility birth.20 Because the financial incentive is given irrespective of whether
the individual would have given birth in the health facility in the absence of the JSY, the
cost per marginal visit is much higher than the value of incentive. Using data on the cost of
delivery from Bonu et al (2009), we calculate a total cost of $415 for each additional facility
birth.21 However, while a cost to the government, one could argue that the financial
incentives should not be considered a cost at all since they represent a transfer of resources.
19 This estimate is calculated by applying the coefficients in column 2 of Table 3 to the respective number of
live births in each set of districts categorised according to the various levels of JSY coverage in 2007/08. 20 This figure is likely to represent a minimum cost since we have not factored in administration of the JSY,
whose economic cost is not captured by programme expenditures. If we assume conservatively that
administration costs represent 10 percent of programme spending, expenditure per additional facility birth was
$321. 21 Bonu and colleagues (2009) report estimates of household expenditure on delivery care from India’s National
Sample Survey in 2004. We use household expenditure on a private facility birth on the basis that this better
reflects the full economic cost of giving birth. Because the public sector is subsidised, expenditure on a public
facility birth is likely to be a gross underestimate. While crude, our cost estimate gives a sense of the order of
magnitude. Note that the financial data are adjusted for inflation.
14
The cost to society then is only the deadweight loss associated with taxation, the
administrative cost of running the JSY and the cost of providing delivery care services.
There is a growing literature on demand-side incentives in health against which to compare
the magnitudes of our estimated effects. In terms of the JSY, we compare our results against
the study by Lim et al (2010) which suggests that the programme increased use of antenatal
care (three visits or more) by 11 percentage points, increased facility births by 44 to 49
percentage points, and reduced neonatal mortality by 2 to 6 deaths per 1,000 live births.22
Clearly our conclusions are much less encouraging with regards to the healthcare utilisation
findings. The estimates of impact on neonatal mortality are of the same order magnitude in
the two studies. The key difference is the mortality effects in Lim et al (2010) are
statistically significant in two of the three identification strategies they use. Beyond the
JSY, there is a strong body of experimental evidence that comes from studies of conditional
cash transfers in Malawi (Baird et al., 2012), Mexico (Fernald et al., 2008; Gertler, 2000;
Gertler, 2004), Nicaragua (Maluccio and Flores, 2005), Brazil (Morris et al., 2004b),
Ecuador (Paxson and Schady, 2008), Honduras (Morris et al., 2004a), one-off financial
incentives in Malawi (Thornton, 2008), and non-financial incentives in India (Banerjee et
al., 2010), although few are specific to maternal health. The interventions in these studies
were targeted towards poor families and most provide some evidence of positive effects on
utilisation of health services and immunization coverage.
V.C Unintended Consequences
Our results thus far have focused on outcomes which, according to the stated objectives of
the programme, the JSY was intended to improve. However, high powered incentives have
the potential to influence a broad range of behaviours, which in turn may have both positive
and negative implications for welfare. Here we study three possible effects of such
incentives. First, we expect the JSY to increase demand for public maternity services, in
part, through a substitution away from private health providers. Second, some have argued
that cash payments for delivery or child health care provide an incentive to become
pregnant. Third, financial incentives for delivery care may have positive benefits through
changes in health-related behaviours subsequent to childbirth, such as breastfeeding. The
idea is that women who give birth in a health facility are more likely to be exposed to
information on the benefits of timely breastfeeding.23
Table 5 presents the results on unintended consequences of the JSY. Column (1) shows
that the JSY was associated with a reduction in utilisation of maternity services in the
private sector. For reference, we reproduce in column (2) previous findings on utilisation
of services in the public sector. Substitution away from the private sector accounts for a
sizeable proportion of the effect of the JSY on public facility births. Data from the DLHS
22 See the Online Appendix for a more detailed comparison of the two set of findings. 23 We also considered other health-related behaviours potentially influenced by exposure to information during
childbirth, including postnatal care seeking, whether the baby was immediately wiped dry and wrapped, and
whether a sterilized blade was used to cut the umbilical cord. The DLHS, however, provides no scope for
measuring these outcomes consistently between the two survey rounds. Child immunization was not regarded
as a plausible indirect outcome given the long time lag between childbirth and vaccinations.
15
lend support to these findings by showing that the JSY has been predominantly a public
sector programme despite the stated policy to involve private health providers. Only 10
percent of JSY beneficiaries nationwide gave birth in a private health facility.
We next look at the results on pregnancies.24 They show in column (4) that the JSY was
associated with a modest increase in the likelihood of a woman being pregnant in a given
year. This result is plausible when we consider that it probably reflects a reduction in birth
spacing rather than an increase in the total lifetime number of children. Either way, there
are implications for health given that both birth spacing and total fertility are important
underlying causes of maternal and neonatal mortality (Zhu et al., 1999). Other studies have
shown that fertility is amenable to change in the face of conditional cash transfers (Morris
et al., 2004a; Stecklov et al., 2006) and cable TV (Jensen and Oster, 2009) and estimates
reported in these studies are much greater than the effect of the JSY found here.
The risk of increased childbearing was partly anticipated by policymakers in the design of
the JSY and these safeguards provide some motivation to scrutinise the validity of the
pregnancy results. If women with more than two children were unable to receive the JSY
cash, why would they be incentivised to become pregnant? However, the policy of limiting
the cash payment to women with two or fewer children applied only to the low focus states
and was difficult to implement. DLHS-3 data show that the probability of a woman
receiving the cash incentive after giving birth in a public health facility is statistically the
same across parity groups, a pattern which suggests policy attempts to mitigate this
unintended consequence were not implemented.25
Columns (5) and (6) of Table 4 report the results on breastfeeding within the first hour and
the first 24 hours of birth, respectively. Both sets of results suggest that the JSY was
associated with an increase in breastfeeding soon after childbirth, consistent with increased
exposure to information from health workers around the time of childbirth.
V.D Robustness
Our estimates of effect are credible in so far as our identifying assumption holds that JSY
coverage is orthogonal to the error term. While it is by definition impossible to test this
assumption formally, we can mitigate concerns of bias due to non-random placement of
the JSY by pursuing several robustness checks. Pre-trends are a commonly used tool to
examine whether the assumption underpinning the difference-in-difference approach is
credible. Specifically, if districts with different levels of coverage of the JSY have similar
24 For women who report being pregnant at the time of interview, we have no information on when they became
pregnant to assign the pregnancy to a specific year. We therefore use a random number generator, constrained
between three and nine, to determine the number of months a woman is pregnant, if pregnant at the time of
interview. The pregnancy for these women is thus assigned to one of two possible years. Another approach
might seek to model seasonality in pregnancy. The data, however, show that the probability of pregnancy differs
little across months of the year. 25 The percentage of women who received the cash incentive conditional on giving birth in a public health
facility is as follows: first birth (33.0 percent); second birth (32.5 percent), third birth (29.1 percent); fourth
birth (33.4 percent); and fifth or higher birth (35.5 percent). While these data are not perfect – the number of
times a woman has given birth does not necessarily equal the number of living children – they are highly
suggestive of the policy not being effective in practice.
16
trends in outcomes prior to the start of the programme, we can be more confident of our
estimates. Descriptive data are reassuring in this respect. Pre-trends plotted separately for
districts in each of the four categories of JSY coverage are similar (see Online Appendix).
More formally, using only data prior to the start of the JSY programme, we examine
whether pre-trends differ according to future JSY coverage. As indicated by the coefficient
on the interaction between years since the start of the data period and future JSY coverage
in Table A5 of the Appendix, we are able to accept at the 5 percent level the null hypothesis
of equal pre-trends.26
We then examine whether JSY coverage is correlated with the characteristics of individual
women. We have argued that JSY coverage is primarily a supply-side measure. This check
provides evidence on whether JSY coverage is correlated with demand once we control for
selection at the policy level. We regress the JSY coverage variable on the full set of
individual-level demographic controls while including the district covariates. Table A6 of
the Appendix presents the results of this robustness check. The results in column (1) are
simply to show that when we fail to account for selection at the district level, individual
demographics are a strong predictor of JSY implementation. An F test of the joint
hypothesis that none of the demographics is correlated with JSY coverage is rejected
(p=0.015). When we do control for selection at the district level, in column (2), we see that
these same demand-side factors are no longer correlated with implementation of the JSY
(p=0.290) despite the fact that, as column (3) shows, they are strong predictors of utilisation
of government delivery care services (p=<0.001). Together these results give us more
confidence that the variation in our measure of JSY coverage is largely supply-driven.
We performed a range of further robustness checks. These are summarised in Table A7 of
the Appendix. Long difference regressions using data at three-year intervals yield
coefficients similar to the main results (Panel A). Dropping districts with a high neonatal
mortality rate of over 50 deaths per 1,000 live births (6 percent of districts) leads to almost
identical coefficients (Panel B). Excluding districts in states where there was no parity
condition connected to the receipt of the cash (ie. states designated by the programme as
low priority) leaves the point estimates essentially the same (Panel C). Finally, allowing
for the possibility of confounding trends, by including state-specific time trends, reduces
the magnitude of the estimates although the general pattern of results remains unchanged
(Panel D).
VI. Heterogeneity in Impacts
We first examine how the effect of the JSY is distributed along several standard dimensions
of socioeconomic status, namely maternal education and household wealth. These can be
considered demand-side factors that may modify the effect of the JSY on health care
seeking behaviour. We then study whether there is a dose-response relationship. By
exploiting the fact that the JSY substantially varies the amount of cash paid to women in
26 These findings hold if we interact time with categories of JSY coverage (result not shown).
17
different places, we are able to learn more about a fundamental policy parameter. The
amount of cash paid to women is more generous in rural areas than urban areas within high
focus states, and in high focus states than low focus states. We therefore conduct two
subgroup analyses along these lines.
Table 6 presents the JSY treatment effects across various subsamples with public facility
births as the dependent variable. The first two columns show that the effect of the JSY on
utilisation is greater amongst women with no education than women with some education
(p value of the difference is <0.001). The next two columns compare the treatment effect
between the two wealth groups, with point estimates showing a similar pattern to the
education results. Poorer women are more likely to give birth in a public health facility in
response to the JSY than richer women (p value of the difference is <0.001). The results in
the remaining four columns show that the effect of the JSY was larger in places where the
amount of cash offered to women was greater. The response to the JSY was greater in rural
areas than urban areas (p value of the difference is 0.058) and greater in low focus than
high focus states (p value of the difference is <0.001). When considered relative to the
baseline mean, the differences between the subgroups are clearly large.
VII. Conclusions
In this paper, we have examined the association between one of the world’s largest demand-
side financial incentive programmes and health-related outcomes in India. Consistent with
much of the literature outside of India, we find that the financial incentives in the JSY are
associated with an increase in the use of formal health services, particularly at lower levels
of the public health system. The increase in use of formal maternal health care due to the
programme was modest. Our findings on neonatal mortality show no strong evidence of an
effect, although confidence intervals are not sufficiently tight to reject modest effects of
the JSY on mortality.
A persuasive explanation for the mortality finding is that the JSY incentivised women
predominantly to health facilities whose purpose was not to manage life-threatening
complications. However good the quality of care in health institutions below the district
hospital, it may remain inadequate to save the lives of women and their baby, particularly
when obstetric emergencies require intensive rather than obstetric care (Costello et al.,
2006). Having a fully functional referral system is thus critical for the success of any
intervention which seeks to increase uptake of institutional delivery care (Campbell and
Graham, 2006). Existing evidence suggests that the quality of maternity services and the
referral system in the public sector remains poor in India (Chaturvedi et al., 2014; Hulton
et al., 2007; Nagpal et al., 2015; Stanton et al., 2014).
We have argued that high powered incentives have the potential to influence a broad range
of behaviours, intended or otherwise. Any evaluation of financial incentives should go
beyond the narrow objectives of the programme to examine potential unintended
consequences. Our pregnancy results are striking because they suggest a pathway through
18
which the programme’s own objective of reducing maternal and neonatal mortality may be
undermined. It also serves to demonstrate the importance of anticipating such risks in the
programme design and, in turn, ensuring appropriate measures are put into practice.
A further point of discussion relates to the generalisability of our findings to an expanded
JSY programme, say five years down the line. It is certainly possible that the effect of the
programme has increased as it has matured. Women will only be incentivised by the
programme if they know about the benefits but it takes time for such information to spread
in the population. Alternatively, the effects in this paper may be larger than those observed
when the JSY finally reaches all districts in India. Early implementation of the JSY was
understandably prioritised in districts that contain poorer populations and evidence on
impact heterogeneity suggests that these districts were the ones where the greatest benefits
from the programme could be realised. Thus, extending our estimates of effect to the period
since 2008 may not provide a good approximation to the true impact of the programme.
The collective evidence in this paper, on both intended and unintended effects, points
towards the need for policymakers to be cautious in the use of financial incentives. For
example, even though it is self-evident that the supply-side must be in place if demand-side
financial incentives are to work, there is a proliferation of schemes in countries where the
quality and even availability of care are vastly inadequate. Future research on this topic
should broaden its scope to address questions around their long-term effects, and the
potential harms they may cause (Lagarde et al., 2007).
19
Table 1. Descriptive Statistics
2001/02-2004/05
(before JSY) 2005/06-2007/08
(during JSY)
Panel A. Study outcomes
Neonatal mortality (per 1,000 live births) 33.0 26.6
One-day mortality (per 1,000 live births) 16.1 13.0
Health worker in attendance at delivery (%) 46.0 49.1
Delivery in a health facility (%) 38.7 43.7
Public health provider (%) 20.0 25.5
Private health provider (%) 18.7 18.2
Caesarean section (%) 7.3 8.1
Assisted delivery (%) 2.6 1.8
At least three antenatal care visits (%) 43.6 46.9
Breastfeeding within one hour of birth (%) 31.1 39.9
Pregnant in a given year (%) - 7.7
Panel B. Individual covariates
Urban (%) 26.1 18.3
Hindu (%) 76.3 76.0
Scheduled caste (%) 18.4 19.0
Scheduled tribe (%) 16.7 17.6
Other backward caste / tribe (%) 40.0 40.6
Maternal age (years) 24.6 25.0
Number of live births 2.64 2.54
Woman’s education (grades completed) 4.36 4.47
Husband’s education (grades completed) 6.66 6.62
Household wealth asset (score) -0.018 -0.053
Notes: Summary statistics are based on data from the DLHS-2 and DLHS-3, including observations over the period 2001/02 – 2007/08. The unit of observation is a woman’s most recent delivery, except in the case of neonatal mortality (live birth) and pregnant this year (woman-year). Assisted delivery includes the use of forceps or a ventouse. The household asset wealth score is generated by applying principal component analysis to a set of household asset ownership variables.
20
Table 2. JSY Coverage
2005/06 2006/07 2007/08
Districts with
JSY coverage 0-10% 279 163 98
JSY coverage 10-25% 151 137 144
JSY coverage 25-50% 123 164 162
JSY coverage >50% 34 123 183
Total sample 587 587 587
Notes: Based on data from the DLHS-3.
21
Table 3. Association of JSY with Use of Maternal Health Care Services
Dependent variable:
Health worker in attendance at delivery
Delivery in a health facility
Delivery in public health facility
Delivery by type of public health facility At least three
ANC visits Hospital Community health centre
Primary health centre
(1) (2) (3) (4) (5) (6) (7)
Panel A. Baseline model
JSY coverage 10-25% -0.0053 -0.0061 -0.00072 0.00081 0.0010 0.0023 0.00071
(0.0055) (0.0054) (0.0049) (0.0044) (0.0017) (0.0019) (0.0053)
JSY coverage 25-50% 0.0017 0.0072 0.019*** 0.0033 0.0092*** 0.011*** 0.0052
(0.0061) (0.0061) (0.0055) (0.0047) (0.0023) (0.0027) (0.0061)
JSY coverage >50% 0.056*** 0.075*** 0.11*** 0.028*** 0.040*** 0.051*** 0.010
(0.0090) (0.0093) (0.0086) (0.0061) (0.0045) (0.0039) (0.0075)
Panel B. Baseline model with district and individual controls
JSY coverage 10-25% -0.0021 -0.0033 -0.0015 -0.00011 0.0016 0.0018 0.00099
(0.0048) (0.0047) (0.0047) (0.0040) (0.0017) (0.0020) (0.0048)
JSY coverage 25-50% 0.0028 0.0075 0.013** 0.00017 0.010*** 0.0078*** 0.0035
(0.0053) (0.0053) (0.0055) (0.0044) (0.0024) (0.0026) (0.0057)
JSY coverage >50% 0.063*** 0.082*** 0.10*** 0.027*** 0.043*** 0.045*** 0.010
(0.0081) (0.0084) (0.0084) (0.0057) (0.0046) (0.0037) (0.0073)
Mean of dep. variable at baseline 0.46 0.39 0.20 0.14 0.02 0.02 0.45
Number of observations 342,875 342,875 342,875 342,875 342,875 342,875 340,323
Notes: Data are from the DLHS-2 and the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors, corrected for clustering at the district level, are reported in parentheses. Health worker is in attendance if the birth is in a health facility or at home with a doctor, nurse, midwife, or lady health volunteer. Baseline model includes fixed effects for district and year of birth. Model with district and individual controls includes interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth and survey round. The unit of observation is a delivery (most recent only). Deviations in sample size are due to missing data.
22
Table 4. Association of JSY with Neonatal Mortality
Dependent variable: Neonatal mortality
Disaggregated measures of mortality
1 day mortality Death between 2 and 28 days Death between 8 and 28 days
(1) (2) (3) (4)
Panel A. Baseline model
JSY coverage 10-25% -0.00078 -0.0013 0.00051 0.00026
(0.0012) (0.00085) (0.00085) (0.00049)
JSY coverage 25-50% -0.00030 -0.00048 0.00018 0.000067
(0.0013) (0.00093) (0.00092) (0.00051)
JSY coverage >50% -0.0031* -0.0020* -0.0011 -0.00057
(0.0016) (0.0012) (0.0011) (0.00065)
Panel B. Baseline model with district and individual controls
JSY coverage 10-25% -0.00043 -0.0012 0.00075 0.00036
(0.0012) (0.00086) (0.00084) (0.00049)
JSY coverage 25-50% 0.00026 -0.00051 0.00077 0.00030
(0.0012) (0.00095) (0.00089) (0.00053)
JSY coverage >50% -0.0027 -0.0022* -0.00053 -0.00029
(0.0017) (0.0012) (0.0011) (0.00066)
Mean of dep. variable at baseline 0.031 0.015 0.016 0.0060
Number of observations 429,443 429,443 429,443 429,443
Notes: Data are from the DLHS-2 and the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors, corrected for clustering at the district level, are reported in parentheses. Baseline model includes fixed effects for district and year of birth. Model with district and individual controls includes interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth and survey round. The unit of observation is a live birth (based on the birth history of a woman).
23
Table 5. Association of JSY with Unintended Outcomes
Dependent variable
Place of delivery Pregnant
(2004 – 2008)
Breastfeeding
Private health facility
Public health facility
Within 1 hour Within 24 hours
(1) (2) (3) (4) (5)
Panel A. Baseline model
JSY coverage 10-25% -0.0053 -0.00072 0.00058 0.016** 0.015*
(0.0041) (0.0049) (0.0011) (0.0071) (0.0082)
JSY coverage 25-50% -0.012*** 0.019*** 0.0011 0.026*** 0.025***
(0.0042) (0.0055) (0.0012) (0.0085) (0.0096)
JSY coverage >50% -0.034*** 0.11*** 0.0070*** 0.075*** 0.076***
(0.0047) (0.0086) (0.0019) (0.011) (0.013)
Panel B. Baseline model with district and individual controls
JSY coverage 10-25% -0.0018 -0.0015 0.0016 0.017** 0.015*
(0.0039) (0.0047) (0.0010) (0.0072) (0.0081)
JSY coverage 25-50% -0.0053 0.013** 0.0026** 0.026*** 0.023**
(0.0040) (0.0055) (0.0012) (0.0087) (0.0097)
JSY coverage >50% -0.022*** 0.10*** 0.0094*** 0.073*** 0.069***
(0.0045) (0.0084) (0.0020) (0.012) (0.014)
Mean of dep. variable at baseline 0.19 0.20 0.086 0.32 0.54
Number of observations 342,875 342,875 2,528,498 336,252 336,252
Notes: Data are from the DLHS-2 and the DLHS-3, except column (3) which uses pregnancy data from women in the DLHS-3 only. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors, corrected for clustering at the district level, are reported in parentheses. Standard deviation of the dependent variable mean is in square brackets. Column (3) assumes that the number of months a woman has been pregnant, if pregnant at the time of interview, is as good as random (constrained to be between three and nine months). Baseline model includes fixed effects for district and year of birth. Model with district and individual controls includes interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth, and survey round. The unit of observation is a delivery (most recent only), except in columns (3) where it is a woman-year.
24
Table 6. Heterogeneity in the Effect of the JSY on Government Facility Births
Education of mother Wealth of household Residence
(in high focus states) Focal states
No education
Some education
Poorest half Richest half Urban Rural High focus Low focus
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A. Baseline model
JSY coverage 0.22*** 0.18*** 0.22*** 0.18*** 0.19*** 0.24*** 0.22*** 0.0083
(0.016) (0.014) (0.017) (0.015) (0.027) (0.017) (0.017) (0.020)
Panel B. Baseline model with district and individual controls
JSY coverage 0.22*** 0.18*** 0.23*** 0.18*** 0.18*** 0.24*** 0.23*** -0.012
(0.016) (0.014) (0.016) (0.015) (0.027) (0.017) (0.016) (0.021)
Mean of dep. variable at baseline 0.11 0.28 0.12 0.28 0.28 0.13 0.16 0.28
Number of observations 161,813 181,062 174,488 168,387 42,155 191,197 233,352 109,523
Notes: Data are from the DLHS-2 and the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors, corrected for clustering at the district level, are reported in parentheses. Standard deviation of the dependent variable mean is in square brackets. Baseline model includes fixed effects for district and year of birth. Model with district and individual controls includes interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth, and survey round. The demographic variable on which the sample is divided is excluded. The unit of observation is a delivery (most recent only).
25
Figure 1. JSY Coverage and Proportion of Women Giving Birth in a Government Facility
26
Appendix
Table A1. District Correlates of JSY Coverage
Wealth Poverty Tribal
population State fixed
effects
Government facility births
at baseline
(1) (2) (3) (4)
Average asset wealth score -0.058*** -0.051*** -0.041*** -0.017** -0.018*
(0.0051) (0.0053) (0.0057) (0.0088) (0.0092)
Poor share of population 0.13*** 0.12** 0.13** 0.13**
(0.032) (0.032) (0.055) (0.056)
Tribal share of population 0.14*** -0.0053 -0.0053
(0.030) (0.037) (0.037)
Government facility share of births 0.0059
(0.044)
State fixed effects No No No Yes Yes
Number of observations 1,761 1,761 1,761 1,761 1,761
Notes: Data are from the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are reported in parentheses. The dependent variable is JSY coverage. The unit of observation is a district-year over the period 2005/06 to 2007/08. Government facility share of births is measured at baseline (2004/05).
27
Table A2. JSY as a Binary Treatment
Dependent variable:
Delivery in a health facility
Delivery in public facility
Delivery in private facility
ANC three visits
Neonatal mortality
One-day mortality
Pregnant (2004-08)
Breastfeeding within 1 hour
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A. Baseline model
JSY coverage > 10% 0.013** 0.025*** -0.012*** 0.0040 -0.00098 -0.0011 0.0014 0.030***
(0.0054) (0.0048) (0.0037) (0.0052) (0.0011) (0.00079) (0.0010) (0.0074)
Panel B. Baseline model with district and individual controls
JSY coverage > 10% 0.013*** 0.019*** -0.0059* 0.0032 -0.00050 -0.0011 0.0025** 0.028***
(0.0046) (0.0047) (0.0035) (0.0048) (0.0011) (0.00080) (0.00099) (0.0074)
Notes: Data are from the DLHS-2 and the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors, corrected for clustering at the district level, are reported in parentheses. Health worker is in attendance if the birth is in a health facility or at home with a doctor, nurse, midwife, or lady health volunteer. Baseline model includes fixed effects for district and year of birth. Model with district and individual controls includes interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth and survey round. The unit of observation is a delivery (most recent only). Deviations in sample size are due to missing data.
28
Table A3. JSY as a Continuous Treatment
Dependent variable:
Delivery in a health facility
Delivery in public facility
Delivery in private facility
ANC three visits
Neonatal mortality
One-day mortality
Pregnant (2004-08)
Breastfeeding within 1 hour
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A. Baseline model
JSY coverage 0.15*** 0.20*** -0.059*** 0.020* -0.0035 -0.0022 0.013*** 0.12***
(0.015) (0.014) (0.0072) (0.012) (0.0024) (0.0018) (0.0030) (0.016)
Panel B. Baseline model with district and individual controls
JSY coverage 0.16*** 0.20*** -0.041*** 0.023** -0.0033 -0.0026 0.017*** 0.12***
(0.014) (0.013) (0.0067) (0.012) (0.0025) (0.0019) (0.061) (0.017)
Notes: Data are from the DLHS-2 and the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors, corrected for clustering at the district level, are reported in parentheses. Health worker is in attendance if the birth is in a health facility or at home with a doctor, nurse, midwife, or lady health volunteer. Baseline model includes fixed effects for district and year of birth. Model with district and individual controls includes interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth and survey round. The unit of observation is a delivery (most recent only). Deviations in sample size are due to missing data.
29
Table A4. Association of JSY with Medical Procedures at Childbirth
Dependent variable: Caesarean section Assisted delivery
(1) (2)
Panel A. Baseline model
JSY coverage 10-25% -0.0038 0.0012
(0.0026) (0.0020)
JSY coverage 25-50% -0.0032 0.0019
(0.0028) (0.0022)
JSY coverage >50% -0.013*** 0.0076***
(0.0032) (0.0027)
Panel B. Baseline model with district and individual controls
JSY coverage 10-25% -0.0021 0.0014
(0.0024) (0.0020)
JSY coverage 25-50% 0.00081 0.0016
(0.0027) (0.0023)
JSY coverage >50% -0.0054* 0.0066**
(0.0031) (0.0028)
Mean of dependent variable at baseline 0.075 0.024
Number of observations 342,853 342,853
Notes: Data are from the DLHS-2 and the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors, corrected for clustering at the district level, are reported in parentheses. Standard deviation of the dependent variable mean is in square brackets. Assisted delivery involves the use of forceps or a ventouse. Baseline model includes fixed effects for district and year of birth. Model with district and individual controls includes interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth, and survey round.
30
Table A5. Differences in Pre-Trends
Delivery in a facility
Delivery in public facility
Neonatal mortality
One-day mortality
(1) (2) (3) (4)
Panel A. Baseline model
Time 0.011*** 0.0042* -0.0046*** -0.0027***
(0.0024) (0.0021) (0.00060) (0.00039)
Time x JSY coverage -0.011* 0.0057 0.00027 0.0013
(0.0064) (0.0058) (0.0017) (0.0010)
Panel B. Baseline model with district and individual controls
Time -0.00019 -0.0086 0.0045*** 0.0014
(0.0056) (0.0064) (0.0017) (0.0012)
Time x JSY coverage -0.00071 0.0090 0.0020 0.0020*
(0.0057) (0.0058) (0.0017) (0.0011)
Mean of dependent variable 0.39 0.20 0.033 0.016
Number of observations 168,887 168,887 226,567 226,567
Notes: Data are from the DLHS-2 and the DLHS-3 but are for the period before the start of the JSY only. *** denotes significance at 1%, ** at 5%, and * at 10% level. Baseline model includes time (birth year since start of data period), an interaction between time and coverage of the JSY, and fixed effects for district. Model with district and individual controls includes interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth, and survey round.
31
Table A6. Correlation between JSY Coverage and Demographics
JSY coverage JSY coverage Delivery in a health facility
(1) (2) (3)
Urban -0.00050 -0.00067 0.049***
(0.00097) (0.00090) -0.0052
Hindu -0.00069 -0.00047 0.030***
(0.00091) (0.00085) -0.005
Scheduled caste -0.00025 0.000036 0.026***
(0.0010) (0.00098) -0.0049
Scheduled tribe 0.00054 0.00034 -0.049***
(0.0013) (0.0012) -0.0073
“Other backward” ethnicity 0.00028 0.00045 0.0034
(0.00082) (0.00076) -0.0045
Woman’s education (grades completed) -0.00016* -0.00014 0.0040***
(0.000093) (0.000086) -0.00043
Husband’s education (grades completed) 0.000028 0.000077 0.0018***
(0.000084) (0.000079) -0.00032
Two live births 0.0030*** 0.0014* -0.048***
(0.00089) (0.00082) -0.0036
Three live births 0.0033*** 0.0025** -0.068***
(0.0011) (0.0011) -0.0046
Four live births 0.0011 0.00080 -0.091***
(0.0013) (0.0012) -0.0052
Five or more live births 0.00063 0.0014 -0.100***
(0.0014) (0.0013) -0.0061
Mother’s age at childbirth (years) -0.000094 -0.00011 0.00018
(0.000085) (0.000079) -0.00032
Wealth asset score 0.00026 0.0000095 -0.0027**
(0.00022) (0.00020) -0.0012
Multiple birth 0.0027 0.0024 0.069***
(0.0028) (0.0026) -0.011
District controls No Yes Yes
F (14, 586) 2.01 1.18 57.72
p-value 0.015 0.290 <0.001
Number of observations 173,988 173,988 173,988
Number of districts 587 587 587
Notes: Data are from the DLHS-2 and the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors, corrected for clustering at the district level, are reported in parentheses. Regressions includes fixed effects for district and year of birth, as well as the variables reported. Regression in column (2) and (3) further include interactions between year of birth and district share of the population below the poverty line, tribal population share, and mean wealth asset score.
32
Table A7. Further Robustness Checks
Delivery in public facility
Delivery in private facility
One-day mortality
Pregnant (2004-08)
Breastfeeding within 1 hour
(1) (2) (3) (4) (5)
Panel A. Three-year long differences (2001, 2004, 2007)
JSY coverage 10-25% -0.0014 0.0028 -0.0020 n/a 0.0057
(0.012) (0.010) (0.0024) (0.019)
JSY coverage 25-50% 0.026** 0.00085 0.00082 n/a 0.0061
(0.012) (0.0100) (0.0025) (0.019)
JSY coverage >50% 0.17*** -0.024** -0.0014 n/a 0.091***
(0.014) (0.0096) (0.0025) (0.019)
Panel B. Exclude high mortality districts
JSY coverage 10-25% -0.00071 -0.0032 -0.00084 0.0018* 0.020***
(0.0050) (0.0042) (0.00081) (0.0011) (0.0075)
JSY coverage 25-50% 0.015** -0.0059 -0.000059 0.0027** 0.026***
(0.0057) (0.0042) (0.00091) (0.0013) (0.0090)
JSY coverage >50% 0.10*** -0.023*** -0.0018 0.0097*** 0.071***
(0.0086) (0.0047) (0.0012) (0.0021) (0.012)
Panel C. Exclude low priority states
JSY coverage 10-25% -0.0016 -0.00027 -0.00060 0.00091 0.026***
(0.0064) (0.0046) (0.0011) (0.0015) (0.0098)
JSY coverage 25-50% 0.017** -0.0039 -0.00060 0.0016 0.046***
(0.0075) (0.0048) (0.0011) (0.0017) (0.012)
JSY coverage >50% 0.11*** -0.021*** -0.0028** 0.0090*** 0.089***
(0.010) (0.0056) (0.0013) (0.0025) (0.015)
Panel D. State-specific time trends
JSY coverage 10-25% -0.0096** -0.0016 -0.0011 0.0029*** 0.013**
(0.0043) (0.0036) (0.00088) (0.0010) (0.0056)
JSY coverage 25-50% -0.010** -0.0017 -0.0011 0.0040*** 0.011
(0.0049) (0.0037) (0.0010) (0.0012) (0.0065)
JSY coverage >50% 0.033*** -0.010** -0.0023* 0.0052** 0.016
(0.0078) (0.0044) (0.0013) (0.0021) (0.010)
Notes: Data are from the DLHS-2 and the DLHS-3. *** denotes significance at 1%, ** at 5%, and * at 10% level. All estimates are from a model that includes fixed effects for district and year of birth, interactions between year of birth and district share of the population below the poverty line, tribal population share, and wealth asset score as well as individual controls for mother’s education, husband’s education, mother’s age at birth, wealth asset score, recall period, and dummies for categories of urban dwelling, religion, number of live births, a multiple birth, and survey round.
33
Online Appendix (Not for Publication)
A. Comparison to Lim et al (2010)
Lim et al (2010) present estimates of the effect of the JSY using the same survey data we use but
different empirical methods. They produce estimates based on three alternative methods of
analysis: exact matching, with versus without, and a difference-in-difference analysis. The first two
methods are based on individual level data while the latter uses data aggregated at the district level.
We compare results to examine whether there are differences. While such a comparison cannot in
itself explain why there are differences, our critical review of the methods used by Lim et al (2010)
in Section II.C suggests that there are some important sources of potential bias in their estimates of
effect.
A comparison of the effect estimates between the two studies is complicated by the fact that
treatment is defined differently. Lim et al (2010) define treatment as whether the woman reports
receiving the JSY cash. In their difference-in-difference analysis, treatment is defined analogously
as the district fraction of births receiving the JSY cash. By contrast, we define treatment as the
district fraction of births in a public health facility in which the woman receives the JSY cash, and
then generate categories of JSY coverage for the empirical analysis.
To make a comparison between the two studies more meaningful we report estimates in which our
treatment variable, JSY coverage, enters the regression as a continuous variable. These are the same
estimates as those reported in Table A3 of the Appendix. While this goes some way to improving
comparability, differences in the definition of treatment between the two studies remain and this
should be kept in mind when interpreting the findings.
Table 1. Comparison with Lim et al (2010)
Lim et al (2010) Powell-Jackson
et al (2015)
Exact matching With versus
without Diff-in-diff
Diff-in-diff
(1) (2) (3) (4)
ANC 3 visits or more (%) 10.7 11.1 10.9 2.3
(9.1 to 12.3) (10.1 to 12.1) (4.6 to 17.2) (0.05 to 4.6)
Facility birth (%) 43.5 43.9 49.2 16.0
(42.5 to 44.6) (43.3 to 44.6) (43.2 to 55.1) (13.6 to 19.0)
Neonatal deaths (per 1,000) -2.3 -2.4 -6.2 -3.3
(-3.7 to -0.9) (-4.1 to -0.7) (-20.4 to 8.1) (-8.3 to 1.6)
Notes: Confidence intervals are reported in parentheses. The treatment effects are in percentage points (ANC 3+, facility births) or deaths per 1,000 (neonatal mortality).
Our findings are much less encouraging than those of Lim et al (2010) in terms of the effects on
healthcare utilisation (Table 1). There is roughly a threefold difference between the two studies in
the effect on facility births and the difference is similar when use of antenatal care is the outcome.
The estimates of impact on neonatal mortality are of the same order magnitude in the two studies.
The main difference lies in the fact that the mortality effects in Lim et al (2010) are statistically
34
significant in two of the three identification strategies they use. The confidence interval around our
negative effect of 3.3 deaths per 1,000 means we are unable to reject a modest effect of the JSY on
neonatal mortality.
B. Pre-trends
Figure 1 to Figure 4 use data prior to the official start of the JSY to show trends for districts
categorised according to different levels of JSY coverage as measured in 2007/08. Pre-trends are
shown for the proportion of women giving birth in a facility, the proportion of women giving birth
in government facility, neonatal mortality per 1,000 live births and one-day mortality per 1,000 live
births. For each outcome, we can see that the trends are similar between different groups of JSY
coverage.
Figure 1. Pre-trends for Facility Births
0.2
.4.6
.81
Fa
cili
ty b
irth
s
2001 2002 2003 2004Year
JSY coverage 0-10% JSY coverage 10-25%
JSY coverage 25-50% JSY coverage >50%
35
Figure 2. Pre-trends for Government Facility Births
Figure 3. Pre-trends for Neonatal Mortality
0.2
.4.6
.81
Go
vern
men
t fa
cili
ty b
irth
s
2001 2002 2003 2004Year
JSY coverage 0-10% JSY coverage 10-25%
JSY coverage 25-50% JSY coverage >50%
0
.02
.04
.06
.08
.1
Ne
on
ata
l m
ort
alit
y
2001 2002 2003 2004Year
JSY coverage 0-10% JSY coverage 10-25%
JSY coverage 25-50% JSY coverage >50%
36
Figure 4. Pre-trends for One-Day Mortality
0
.02
.04
.06
.08
.1
On
e-d
ay m
ort
alit
y
2001 2002 2003 2004Year
JSY coverage 0-10% JSY coverage 10-25%
JSY coverage 25-50% JSY coverage >50%
37
References
Babiarz KS, et al. New evidence on the impact of China's New Rural Cooperative Medical Scheme and
its implications for rural primary healthcare: multivariate difference-in-difference analysis. BMJ
2010;341; c5617.
Baird SJ, et al. Effect of a cash transfer programme for schooling on prevalence of HIV and herpes
simplex type 2 in Malawi: a cluster randomised trial. Lancet 2012;379; 1320-1329.
Banerjee AV, et al. Improving immunisation coverage in rural India: clustered randomised controlled
evaluation of immunisation campaigns with and without incentives. BMJ 2010;340; c2220.
Bhutta ZA, et al. What works? Interventions for maternal and child undernutrition and survival. Lancet
2008;371; 417-440.
Bhutta ZA, et al. Countdown to 2015 decade report (2000-10): taking stock of maternal, newborn, and
child survival. Lancet 2010;375; 2032-2044.
Bonu S, et al. Incidence and correlates of 'catastrophic' maternal health care expenditure in India. Health
Policy Plann 2009;24; 445-456.
Campbell OM, Graham WJ. Strategies for reducing maternal mortality: getting on with what works.
Lancet 2006;368; 1284-1299.
Chaturvedi S, et al. Quality of obstetric referral services in India's JSY cash transfer programme for
institutional births: a study from Madhya Pradesh province. PloS one 2014;9; e96773.
Chaudhury N, et al. Missing in action: teacher and health worker absence in developing countries. J
Econ Perspect 2006;20; 91-116.
Costello A, et al. An alternative strategy to reduce maternal mortality. Lancet 2006;368; 1477-1479.
Darmstadt GL, et al. Evidence-based, cost-effective interventions: how many newborn babies can we
save? Lancet 2005;365; 977-988.
Das J, Hammer J. 2006. The Quality of Medical Care in India. In: Basu K (Ed)^(Eds), The Oxford
Companion to Economics in India. Oxford University Press: New Delhi, India; 2006.
Das J, Hammer J. Money for nothing: The dire straits of medical practice in Delhi, India. J Dev Econ
2007;83; 1-36.
Das J, et al. The quality of medical advice in low-income countries. J Econ Perspect 2008;22; 93-114.
Deaton A. Instruments, Randomization, and Learning about Development. J Econ Lit 2010;48; 424-
455.
Devadasan N, et al. 2008. A conditional cash assistance programme for promoting institutional
deliveries among the poor in India: process evaluation results. In: Richard F, Witter S, de Brouwere
V (Ed)^(Eds), Reducing financial barriers to obstetric care in low-income countries. ITM: Antwerp;
2008.
Dugger C. 2004. To Help Poor Be Pupils, Not Wage Earners, Brazil Pays Parents. (Ed)^(Eds), New
York Times. New York City; 2004.
Fernald LC, et al. Role of cash in conditional cash transfer programmes for child health, growth, and
development: an analysis of Mexico's Oportunidades. Lancet 2008;371; 828-837.
Finkelstein A, et al. 2011. The Oregon Health Insurance Experiment Evidence from the First Year.
(Ed)^(Eds), NBER working paper series no. w17190. National Bureau of Economic Research:
Cambridge, Mass.; 2011.
Fiszbein A, Schady N. Conditional cash transfers: reducing present and future povertys. World Bank:
Washington DC; 2009.
38
Gertler P. 2000. Final report: the impact of PROGESA on health. (Ed)^(Eds). International Food Policy
Research Institute: Washington DC; 2000.
Gertler P. Do conditional cash transfers improve child health? Evidence from PROGRESA's control
randomized experiment. Am Econ Rev 2004;94; 336-341.
Gertler P, Van der Gaag J. The willingness to pay for medical care: evidence from two developing
countriess. Johns Hopkins University Press: Baltimore and London; 1990.
Hulton LA, et al. Applying a framework for assessing the quality of maternal health services in urban
India. Soc Sci Med 2007;64; 2083-2095.
Human Rights Watch. No Tally of Anguish: Accountability of Maternal Health Care in Indias. Human
Rights Watch: New York; 2009.
Hunter BM, et al. Demand-side Financing and Promotion of Maternal Health: What Has India Learnt?
Economic and Political Weekly 2014;159.
Imbens GW, Wooldridge JM. Recent Developments in the Econometrics of Program Evaluation. J Econ
Lit 2009;47; 5-86.
International Institute for Population Sciences. National Family Health Survey (MCH and Family
Planning), India 1992-93s. IIPS: Mumbai; 1995.
International Institute for Population Sciences. District Level Household and Facility Survey (DLHS-
3), 2007-08: Indias. IIPS: Mumbai; 2010.
International Institute for Population Sciences, Macro International. National Family Health Survey
(NFHS-3), 2005–06: India: Volume Is. IIPS: Mumbai; 2007.
Jeffery P, Jeffery R. Only when the boat has started sinking: a maternal death in rural north India. Soc
Sci Med 2010;71; 1711-1718.
Jensen R, Oster E. The Power of TV: Cable Television and Women's Status in India. Q J Econ 2009;124.
Jones G, et al. How many child deaths can we prevent this year? Lancet 2003;362; 65-71.
Kassebaum NJ, et al. Global, regional, and national levels and causes of maternal mortality during 1990-
2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet 2014.
King G, et al. Public policy for the poor? A randomised assessment of the Mexican universal health
insurance programme. Lancet 2009;373; 1447-1454.
Lagarde M, et al. Conditional cash transfers for improving uptake of health interventions in low- and
middle-income countries: a systematic review. JAMA 2007;298; 1900-1910.
Lim SS, et al. India's Janani Suraksha Yojana, a conditional cash transfer programme to increase births
in health facilities: an impact evaluation. Lancet 2010;375; 2009-2023.
Malini S, et al. A rapid appraisal on functioning of Janani Suraksha Yojana in South Orissa. Health and
Population: Perspectives and Issues 2008;31; 126-131.
Maluccio J, Flores R. 2005. Impact evaluation of a conditional cash transfer program: the Nicaraguan
Red de Protección Social. (Ed)^(Eds). International Food Policy Research Insitute: Washington DC;
2005.
Manning WG, et al. Health-Insurance and the Demand for Medical-Care - Evidence from a Randomized
Experiment. Am Econ Rev 1987;77; 251-277.
Marteau TM, et al. Using financial incentives to achieve healthy behaviour. BMJ 2009;338; b1415.
Mills A, et al. What do we mean by rigorous health-systems research? Lancet 2008;372; 1527-1529.
Ministry of Health and Family Welfare. 2005. National Rural Health Mission: Mission Document.
(Ed)^(Eds). Government of India: Delhi; 2005.
Ministry of Health and Family Welfare. 2006. Janani Suraksha Yojana: features & frequently asked
questions and answers. (Ed)^(Eds). Government of India: New Delhi; 2006.
39
Morris S, et al. Monetary incentives in primary health care and effects on use and coverage of preventive
health care interventions in rural Honduras: cluster randomised trial. Lancet 2004a;364; 2030-2037.
Morris SS, et al. Conditional cash transfers are associated with a small reduction in the rate of weight
gain of preschool children in northeast Brazil. J Nutr 2004b;134; 2336-2341.
Muralidharan K, et al. 2011. Is There a Doctor in the House? Medical Worker Absence in India.
(Ed)^(Eds). 2011.
Murray SF, et al. Effects of demand-side financing on utilisation, experiences and outcomes of maternity
care in low- and middle-income countries: a systematic review. BMC pregnancy and childbirth
2014;14; 30.
Nagpal J, et al. Widespread non-adherence to evidence-based maternity care guidelines: a population-
based cluster randomised household survey. BJOG : an international journal of obstetrics and
gynaecology 2015;122; 238-247.
Paxson C, Schady N. 2008. Does money matter? The effects of cash transfers on child health and
development in rural Ecuador. (Ed)^(Eds). World Bank: Washington DC; 2008.
Registrar General of India. 2006. Maternal Mortality in India: 1997-2003: Trends, Causes And Risk
Factors. (Ed)^(Eds). Office of the Registrar General of India: New Delhi; 2006.
Registrar General of India. 2013. Special Bulletin on Maternal Mortality in India: 2010-12. (Ed)^(Eds).
Office of the Registrar General of India: New Delhi; 2013.
Rohde J, et al. 30 years after Alma-Ata: has primary health care worked in countries? Lancet 2008;372;
950-961.
Sood N, et al. Government health insurance for people below poverty line in India: quasi-experimental
evaluation of insurance and health outcomes. BMJ 2014;349; g5114.
Stanton CK, et al. Direct observation of uterotonic drug use at public health facility-based deliveries in
four districts in India. International journal of gynaecology and obstetrics: the official organ of the
International Federation of Gynaecology and Obstetrics 2014;127; 25-30.
Stecklov G, et al. Demographic externalities from poverty programs in developing countries:
experimental evidence from Latin Americas. American University Department of Economics:
Washington DC 2006.
Thornton RL. The Demand for, and Impact of, Learning HIV Status. Am Econ Rev 2008;98; 1829-
1863.
Thornton RL, et al. Social security health insurance for the informal sector in Nicaragua: a randomized
evaluation. Health Econ 2010;19 Suppl; 181-206.
UNFPA. Concurrent Assessment of Janani Suraksha Yojana in Selected States – Bihar, Madhya
Pradesh, Orissa, Rajasthan and Uttar Pradeshs. UNFPA: New Delhi; 2009.
United Nations. Global Strategy for Women's and Children's Healths. United Nations, Office of the
Secretary General: New York; 2010.
Verma D, et al. Increasing Institutional Delivery and Acess to Emergency Obstetric Care Services in
Rural Uttar Pradesh. The Journal of Family Welfare 2010;56; 23-30.
Wagstaff A, et al. Extending health insurance to the rural population: an impact evaluation of China's
new cooperative medical scheme. J Health Econ 2009;28; 1-19.
Zhu BP, et al. Effect of the interval between pregnancies on perinatal outcomes. N Engl J Med 1999;340;
589-594.